,4,
SSC-346
FATIGUE CHARACTERIZATION
OF FABRICATED SHIP DETAILS
(Phase 2)
This document has been approved
for public release and sale; its distribution is unlimited
SHIP STRUCTURE COMMITTEE
SHIP STRUCTURE COM1IUEE
The SHIP STRUCTURE COMMI I hE is constituted to prosecute a research program to improve the hull structures of ships and other marine structures by an extension of knowledge pertaining to design, materials, and methods of construction.
RADM J. D. Sipes, USCG, (Chairman)
Chief, Office of Marine Safety, Security and Environmental Protection
U. S. Coast Guard
Mr. Alexander Malakhofl Director, Structural Integrity
Subgroup (SEA 55Y)
Naval Sea S'stoms Command
Dr. Donald Llu
Senior Vice President American Bureau of Shipping
CONTRACTING OFFICER TECHNICAL REPRESENTATIVES Mr. William J. Siekierka
SEA 55Y3
Naval Sea Systems Command
SHIP STRUCTURE SUBCOMMITTEE
The SHIP STRUCTURE SUBCOMMITTEE acts for the Ship Structure Committee on technical matters by providing technical coordination for determinating the goals and objectives of the program and by evaluating and interpreting the results in terms of structural design, construction, and operation.
AMRlCAN;U- _
."I
Mr. Stephen G. Arntson (Chairman) Mr. John F. Conlon
Mr. William Hanzalek Mr, Philip G. Rynn
MILITARY SEALIFT COMMAND
Mr. Albert J. Attermeyer Mr. Michael W. Tourna Mr. Jeffery E. Beach MARITIME ADMINISTRATION Mr. Frederick Seibold Mr. Norman O. Hammer Mr. Chao H. Lin Dr. Walter M. Maclean
SHIP STRUCTURE SUBCOMMITTEE LIAISON MEMBERS U. S. COAST GUARD ACADEMY
LT Bruce Mustain
LI, S. MERCHANT MARINE AÇJ,DEMY
Dr. C. B. Kim
U.S. NAVAL ACADEMY Dr. Ramswar Bhattacharyya
STATE UNIVERSITY OF NEW YORK MARITIME COLLEGE
Dr. W. R. Porter
WELDING RESEARCH COUNCIL
Mr. H. T. Haller
Associate Administrator for
Ship-building and Ship Operations
Maritime Administration
Mr. Thomas W. AlIen Engineering Officer (N7)
Military Sealift Command
CDR Michael K. Parmelee, USCG, Secretary, Ship Structure Committee
U. S. Coast Guard
Mr. Greg D. Woods SEA 55Y3
Naval Sea Systems Command
NAVAL S.EA SYSIEMS COMMANQ Mr. Robert A. Sielski Mr. Charles L. Null Mr. W. Thomas Packard Mr. Allen H. Engle U. S. COAST GUARD CAPT T. E. Thompson
CAPT Donald S. Jensen
CDR Mark E. NOII
NATIONAL ACADEMY OF SCIENCES
-MARIN E BOARD
Mr. Alexander D. Stavovy
NATIONAL ACADEMY OF SCIENCES -COMMITTEE ON MARINE STRUCTURES Mr. Stanley G. Stiansen
SOCIETY OF NAVAL ARCHITECTS AND MARINE ENGINEERS
-HYDRODYNAMICS COMMITTEE Dr, William Sandberg
AMERICAN IRON AND STEEL INSTITUTE Mr. Alexander D. Wilson
Member Agencies:
United States Coast Guard Naval Sea Systems Command Maritime Administration Amercan Bureau of Shipping Military Sealift Command
Ship
Structure
Cornmittee
An Interagency Advisory Committee
Dedicated to the Improvement of Manne Structures
December 3, 1990
FATIGUE CHARACTERIZATION OF FABRICATED
SHIP DETAILS (PHASE 2)
A basic understanding of fatigue characteristics of fabricated
details
is necessary to ensure the continued reliability and
safety of
ship structures.
Phase
1of this study (SSC-318)
provided a fatigue design procedure for selecting and evaluating
these details.
In this second phase,
an extensive series of
fatigue tests were carried on structural details using variable
loads
tosimulate
avessel's
service history.
This
report
contains the test results as well as fatigue predictions obtained
from available analytical models.
Address Correspondence to:
Secretary, Ship Structure Committee U.S. Coast Guard (G-Mm)
2100 Second Street S.W. Washington, D.C. 20593-0001 PH: (202) 267-0003 FAX: (202) 267-0025 SSC- 34 6 SR- 1298
J) SIPES
Rear Admiral, U.S. Coast Guard
Technical Report Documentation Page 1. R.pert No.
SSC-346
2. Gor.rnm.ne Acc.ss,en N0. 3. R.cip..nts Catalog No.
1. TitI. and Subtitl
Fatigue Characterization of Fabricated Ship Details for Design - Phase II
5. Report Dote
May 1988
6. P.rforming Organiiat,orr Cod.
Ship Structure Committee
8. P.rforrning Orgonizotion R.port No.
SR- 12 98
7. Author's)
S. K. Park and F. V. Lawrence
9. Performing Orgen.ze?,en Nom. end Address
Department of Civil Engineering
University of Illinois at Urbana-Champaign
205 N. Mathews Avenue
Urbana, IL 61801
10. Work Uni? No. (TRAIS)
II. Contract or Grant No.
DTCG 23-84-C-20018
13. lype f Report end P.riod C..r.d
Final Technical Report
12. Sponsorng Agency Nome and Ad.ss
U. S. Coast Guard 2100 2nd Street S.W.
Washington, DC 20593 14. Sponsoring Agency Cod.
G-M
IS. Suppl.n'entory Notes
The U.S.C.G. acts as the contracting office for the Ship Structure Committee.
16. Abstract
The available analytical models for predicting the fatigue behavior of
weldinents under variable amplitude
load histories were compared using test results for weidments subjected to the SAE bracket and transmission variable
load amplitude histories. Models based on detail S-N diagrams
such as the Munse
Fatigue Design Procedure (MFDP) were found to perform well except when the
history had a significant average mean stress.
Models based on fatigue crack propagation alone were generally conservative, while a model based upon estimates of both fatigue crack initiation
and propagation (the I-P Model) performed the best.
An extensive series of fatigue tests was carried out on welded
structural
details commonly encountered in ship construction using a variable load history which simulated the service history of a ship.
The results from this study
showed that linear cumulative damage concepts predicted the
test results, but the
importance of small stress range events was not studied because
events smaller
than 68 MPa (10 ksi) stress range were deleted from the developed
ship history to
reduce the time required for testing.
An appreciable effect of mean stress was observed but the results did not verify the existence of a specimen-size effect.
Baseline constant-amplitude S-N diagrams were developed for five
complex
ship details not commonly studied in the past.
17. Key Words
Fatigue, Ship Structure Details, Design, Reliability, Loading History, Variable Load Histories
18. Distribution Stat.,nenl
Document is available to the U.S. public, the National Technical Information Service,
Springfield, VA 22151
19. Security Clossif. (et thu report)
Unclassified
- 20. Security Classif. (of this page)
Unclassified
21. No. of Pagea
201
22. Price
Appisilmils C.av.iskas I. Mstik Musai..
WI.. Y.. 5Mw
Multiply by
le Pld
$S1
METRIC CONVERSION FACTORS
M
a
-R MAIS (wsight) _____E
sS 25 __________.
5.03 hiI.,uE
_tMs. s., iss... t -125501W a . VOLUME WSIpOs.5 I )IIlIi4«p sI ubI..ss.i il s.lIiIiis.i Ml -fluid a.c.0 21 iÑIlilil.,. M . C15 5.24 IÓIS(S IE
i. pat. .4lis
I -qNsiti 5.21 Ill.i. I øslIsss 3.5 lii«. I -__ a CubiC Isst 0.03 CubiC .41511 Ui3 cubic y.id. 5.75 CubiC 5515v.a
-TEMPERATURE (es.ct)______
-FIkiSidiSit III (stt Culsiul=
M tI51Çela4«S Iuh.òClIil tsispstltas -32) IIZI ..j.II3. I.. 011M oau
CWI.,I5J lI.C. e. 315$ hic. P,ci.
S6,
ol *.,lii - Me.i.11. Pvc. 12.2$. SO Cuijiul N... C13.IO:255.
App,.xi.ale C.avets4e.s is. Mehl. M..s.i.s
$1
WI.. V.. Ssaa Multiply by T. Had SVUISI LINSTN AIEA Cpi' 5 CpR4h.4N5 SIS 54 I.u5 IU 2 s., isvsii
s,ICM ys.s Is.' s hlI..IS($ 5.4 be h.cs. (IS SOS s( 2.1 c MASS (walsh s 0.s* C.. SI kltsCMa 2.2pis
t Ms... lION kil 1.1 OLUMI 5) .IIlilit.v. 0.53 huid ss.s.s It sc I lIess LI paw p' I lIsis IN spsisi I huis 0.21 NIls.. as) Cubic 541515 21 cubic 1551 fI cubic .411«s 1.3 cubIc ylids pl3 1.3 cubIc ylids pl3TEMPERATURE I....)TEMPERATURE I....)
Calsius S/I (di.. habisiMsil P
t«as
.
12) siMiN5 Calsius S/I (di.. habisiMsil Pt«as
.
12) siMiN5 31 S5 -40 0 (40 50 ISO ISO SOOL b -40 -50 50 50 O I 31 31 S5 -40 0 (40 50 ISO ISO SOOL b -40 -50 50 50 O I 31 _IlI_l.Ii U.N _IlI_l.Ii U.Nì.s.
biì.s.
bi CM CISliMSISI. 0,4 CM CISliMSISI. 0,4 isiMs bi isiMs bi M s4«s 3.3 M s4«s 3.3 I. U I. U S 544.45 S 544.45Ii
VS$Ii
VS$ h. hil«wv, h. hil«wv, 0.5 5.I.11 0.5 5.I.11TABLE OF CONTENTS
Page
EXECUTIVE SUMMARY ix
LIST OF SYMBOLS x
INTRODUCTION AND BACKGROUND 1
1.1 The Fatigue Structural Weldments 1
1.2 The Fatigue Design of Weldxnents 1
1.3 Factors Influencing the Fatigue Life of Weidments 2
1.4 Purpose of the Current Study 3
1.5 The Munse Fatigue Design Procedure (MFDP) 4
1.6 References 6
Table 7
Figures 8
COMPARISON OF THE AVAILABLE FATIGUE LIFE
PREDICTION METHODS (TASK 1) 15
2.1 Models Based on S-N Diagrams 15
2.2 Methods Based upon Fracture Mechanics 18
2.3 Comparisons of Predictions with Test Results 20
2.4 References 21
Tables 23
Figures 25
FATIGUE TESTING OF SELECTED SHIP STRUCTURAL DETAILS
UNDER A VARIABLE SHIP BLOCK LOAD HISTORY (TASKS 2-4) 35
3.1 Determination of the Variable Block History 35
3.2 Development of a "Random" Ship Load History 37
3.3 Choice of Detail No. 20 and Specimen Design 38
3.4 Materials and Specimen Fabrication 38
3.6 Test Results and Discussion 42
3.7 Task 2 - Long Life Variable Load History 42
3.8 Task 3 - Mean Stress Effects 42
3.9 Task 4 - Thickness Effects 44
3.10 References 47
Tables 48
Figures 64
FATIGUE LIFE PREDICTION (TASK 6) 86
4.1 Predictions of the Test Results Using the MFDP 86
4.2 Predictions of the Test Results Using the I-P Model 87
4.3 Modeling the Fatigue Resistance of Weldents 88
Tables 93
Figures 99
FATIGUE TESTING 0F SHIP STRUCTURAL DETAILS
UNDER CONSTANT AMPLITUDE LOADING (TASK 7) 112
5.1 Materials and Welding Process 112
5.2 Specimen Preparation, Testing Conditions and Test Results . 112
5.2.1 Detail No. 34 - A Fillet Welded Lap Joint 112
5.2.2 Detail No. 39-A - A Fillet Welded I-Beam
With a Center Plate Intersecting the Web
and One Flange 113
5.2.3 Detail No. 43-A - A Partial-Penetration Butt Weld 114
5.2.4 Detail No. 44 - Tubular Cantilever Beam 114
5.2.5 Detail No. 47 - A Fillet Welded Tubular Penetration 115
Tables 116
Figures 122
SUMMARY AND CONCLUSIONS 151
Page 6.2 The Use of Linear Cumulative Damage (Task 2)
152 6.3 The Effects of Mean Stress (Task 3)
152
6.4 Size Effect (Task 4)
153 6.5 Use of the I-P Model as a Stochastic Model (Task 5)
153 6.6 Baseline Data for Ship Details (Task
7) 154
6.7 Conclusions
154
7. SUGGESTIONS FOR FUTURE STUDY
156
APPENDIX A ESTIMATING THE FATIGUE LIFE OF WELDMENTS USING THE IP MODEL
157 A-1 Introduction
157
A-2 Estimating the Fatigue Crack
Initiation Life (N1) 158
A-2.1 Defining the Stress History (Task 1)
158
A-2.2 Determining the Effects of
Geometry (Task 2) . . 159
A-2.3 Estimating the Residual Stresses (Task 3)
160
A-2.4 Material Properties (Tasks 4-6)
161
A-2.5 Estimating the Fatigue Notch
Factor (Task 7) 163
A-3 The Set-up Cycle (Task 8)
164 A-4 The Damage Analysis (Task 10)
166
A-4.1 Predicting the Fatigue Behavior Under Constant Amplitude Loading With No Notch-Root Yielding or
Mean-Stress Relaxation
166
A-4.2 Predicting the Fatigue Behavior Under Constant Amplitude Loading With Notch-Root Yielding and No Mean-Stress Relaxation
170
A-4.3 Predicting the Fatigue Crack
Initiation Life
Under Constant Amplitude Loading with
Notch-Root Yielding and Mean-Stress Relaxation . . . . 170
A-4.4 Predicting the Fatigue Crack
Initiation Life
Under Variable Load Histories Without Mean Stress Relaxation
A-5 Estimating the Fatigue Life Devoted to
Crack Propagation (Ne) 172
Figures 176
APPENDIX B DERIVATION OF THE MEAN STRESS AND THICKNESS
CORRECTIONS TO THE MIJNSE FATIGUE DESIGN PROCEDURE 196
Figure 199
APPENDIX C SCHEMATIC DESCRIPTION OF SAE BRACKET AND
TRANSMISSION HISTORIES 200
EXECUTIVE SUMMARY
This study is a continuation of a research effort at the University of Illinois at Urbana-Champaign (UIUC) to characterize the fatigue behavior of fabricated ship details. The current study evaluated the Munse Fatigue
Design Procedure and performed further tests on ship details.
The available analytical models for predicting the fatigue behavior of weidments under variable amplitude load histories were compared using test results for weidments subjected to the SAE bracket and transmission variable load amplitude histories. Models based on detail S-N diagrams such as the Munse Fatigue Design Procedure (MFDP) were found to perform well except when the history had a significant average mean stress. Models based on fatigue crack propagation alone were generally conservative, while a model based upon estimates of both fatigue crack initiation and propagation (the I-P Model) performed the best.
An extensive series of fatigue tests was carried out on welded
struc-tural details commonly encountered in ship construction using a variable load history which simulated the service history of a ship. The results from this study showed that linear cumulative damage concepts predicted the test results, but the importance of small stress range events was not
studied because events smaller than 68 MPa (10 ksi) stress range were deleted from the developed ship history to reduce the time required for
testing. An appreciable effect of mean stress was observed, but the results
did not verify the existence of a specimen-size effect.
Both the Munse Fatigue Design Procedure (MFDP) and the I-P Model were
used to predict the test results. The MFDP predicted the mean fatigue life
reasonably well. Improved life predictions were obtained when the effect of
mean stress was included in the MFDP. Mean stress and detail size correc-tions were suggested for the Munse Fatigue Design Procedure.
Generally good results were obtained using the I-P Model, but the
predictions for the smallest size weldments were very unconservative. The I-P model was used to develop a stochastic model for weldment fatigue
behavior based on the observed random variations in specimen geometry and induced secondary stresses resulting from distortions produced by welding. Design aids based on the I-P model are presented.
Baseline constant-amplitude S-N diagrams were developed for five
complex ship details not commonly studied in the past.
a, a., af crack length, initial crack length, final crack length
b fatigue strength exponent
C fatigue crack growth coefficient; also, S-N curve coefficient
c fatigue ductility exponent; also, original half length of
incomplete joint penetration and length of major axis of elliptical crack
D., Dblk
fatigue damage per cycle and per block, respectivelyE Young's Modulus
F function of residual stress distribution
IJP incomplete joint penetration
K' cyclic strength coefficient
K , K fatigue notch factor and maximum fatigue notch factor
f fmax
Kt elastic stress concentration factor
Krms, KlS max min K r M M s' t' LIST OF SYMBOLS
maximum and minimum root mean square stress intensity factor
stress intensity factor due to residual stress
AK stress intensity factor range
AK root mean square stress intensity factor range
rms
k the Weibull scale or shape parameter
Ll leg length of weld perpendicular to the IJP
L2 S-N curve slope parallel to the IJP
magnification factor for free surface, width and stress
gradient
m reciprocal slope of the S-N diagram
Nf cycles to failure
2N - reversals to failure fi
N., Nfi cycles to failure at ith amplitude
n fatigue crack growth exponent; also, thickness effect
exponent
n' cyclic strain hardening exponent
ni cycles at ith amplitude
R stress ratio
RF reliability factor
r notch root radius
S, Sa remote stress range, amplitude of remote stress
remote axial stress amplitude
remote bending stress amplitude
5rms 5rms
stress and root mean square of maximum and minimum stress max min
total remote stress amplitude
Sm applied average axial mean stress
gripping bending stress
S ultimate tensile stress
u
t plate thickness; also, time
w specimen width
x ratio of remote bending stress amplitude to total remote
stress amplitude
X ratio of K to KA
max max
Y geometry factor on stress intensity factor
y(t) stress (strain) spectral ordinate for random time load history
z ratio of bending stress to axial stress
a, ß, A geometry coefficients for elastic stress concentration factor
Ef possible error in fatigue model
Sf coefficient of variation in fatigue life
LSD maximum allowable design stress range expected once during the entire life of a structure
maximum allowable design stress range expected once during
m
the entire life of a structure with allowance for baseline data and applied history mean stress
average constant amplitude fatigue strength at the desired
design life
SN R( average constant amplitude fatigue strength at the desired
design life from baseline testing conditions under stress ratio (R) = R
SN
tSN(l) average constant amplitude fatigue strength at the desired
design life from R = -1 baseline testing conditions
local strain and local strain range
fatigue ductility coefficient
e strain normal to the crack n
spectral ordinates output from the FFT analyzer (in volts)
random phase angle sampled from a uniform distribution, 02ir
a, M, a
local stress, local stress range, local stress amplitudea , a mean stress and residual stress
r
local maximum principal stress amplitude
¿
effective residual stress amplitude fatigue strength coefficientflank angle of welds
random load factor
uncertainty in the mean intercept of the S-N repression line
uncertainty in the fatigue data life
total uncertainty
1. INTRODUCTION AND BACKGROUND
1.1 The Fatigue Structural Weldments
Ships, like most other welded steel structures which are subjected to
fluctuating loads, are prone to metallic fatigue. While fatigue can occur
in any metal component, weldments are of particular concern because of their wide use, because they provide the stress concentrators and, because they
are, therefore, likely sites for fatigue to occur. It is for these reasons
that the fatigue of weldutents has been so exhaustively studied. However,
despite 100 years of research and thousands of studies of weldment fatigue,
there seems to be only slow progress in putting this problem to rest. This slow progress is probably due to the following:
There is a nearly infinite variety of welded joints.
Weldinents of the same j oint type are usually not exactly alike.
The behavior of even simple weldments can be exceedingly complex. The stresses in a weldment are usually imprecisely known.
The variety and complexity of the more common structural weldments are
evi-dent in Fig l-1 which shows the structural details covered in the AISC
fa-tigue provisions [l-l].
1.2 The Fatigue Design of Weldments
There are three main approaches to the fatigue design of weldments:
S-N diagrams Weldinents may be designed using the S-N curves for the
particular detail. The behavior of weldznents under constant amplitude
load-ing has been reported in the literature for hundreds of different j oint
geometries. Attempts to collect the available information and develop a
weidment fatigue data base have been undertaken at the University of
Illinois by Munse [l-2] and by The Welding Institute [1-3]. A typical
collection of weldznent fatigue data from the University of Illinois Data Bank is shown in Fig. l-2 in which it is evident that the fatigue resistance of low stress concentration fatigue-efficient weidments is less than plain
plate and is characterized by a great deal of scatter. Munse [l-4] proposed a fatigue design procedure which uses the "baseline" S-N diagram information
both the desired level of reliability and the variable nature of the applied
loads (Fig. l-4). A short description of the Munse Fatigue Design Procedure
is given in Section 1-5.
Fracture Mechanics: Because fatigue is a process which begins at
stress concentrations (notches), several analytical methods of weldment
fatigue design have recently been developed which are based on mechanics
analyses of fatigue crack initiation and fatigue crack growth at the
critical locations in the structure. Such design methods or analyses
involve sophisticated, complex models (see Fig. l-4). Models based on both
fatigue crack initiation and growth have been proposed by Lawrence et al.
[l-5]: see Appendix A. Models based on fatigue crack growth alone have been
suggested by Maddox [l-6] and Shilling, et al. [l-71.
Structural Tests: A third alternative for the fatigue design of
struc-tures is to base the design on full-scale tests or observations of service
history. While such observations are closest to reality, full-scale tests
are usually prohibitively expensive and time consuming. Moreover, it is sometimes difficult to apply results from one structure to another. In the
case of ships, such tests may require a 20 year study.
1.3 Factors Influencing the Fatigue Life of Weidments
There are four attributes of weldments which, together with the magni--tude of the fluctuating stresses applied, determine the slope and intercept
of their S-N diagram: the ratio of the applied or self-induced axial and bending stresses; the severity of the discontinuity or notch which is an
inherent property of the geometry of the joint; the notch-root residual
stresses which result from fabrication and subsequent use of the welclment,
and the mechanical properties of the material in which fatigue crack
initia-tion and propagainitia-tion take place. 0f these four, the mechanical properties
are probably the least influential.
In most engineering design situations involving as-welded weldments of
a given material, the permissible design stresses are governed by: the
joint geometry, the desired level of reliability, the variable nature of_-the
indication of the sensitivity of the fatigue design stress to these design
variables. The design stress varies greatly with detail geometry, desired
level of reliability and the nature of the variable load. Mean stress has
only a modest influence.
I 1.4 Purpose of the Current Study
This report summarizes a research program sponsored by the U.S. Coast
Guard at the University of Illinois at Champaign-Urbana on the "Fatigue
Characterization of Fabricated Ship Details, Phase II" (contract DTCG
23-84-C-20018). This program is a continuation of one begun at the University of
Illinois under the direction of Professor W. H. Munse [l-4]. The second
j phase had as its principal objectives:
* To evaluate the Munse Fatigue Design Procedure developed and
dis-cussed under Phase I of the project;
* To carry out laboratory fatigue tests of fabricated ship details; * And to perform further tests on ship details.
The tasks of this study are summarized in Table 1-l.
Seven tasks were originally proposed, and they may be broken into four
categories: The first category, Task 1 was a comparison of the Munse
Fatigue Design Procedure (MFDP) predictions with the predictions resulting
from other methods of estimating the fatigue life of weldments and an
assessment of the accuracy of the Munse Fatigue Design Procedure in general.
The results of this comparison are summarized in Section 2.
In the second category, Tasks 2-4 involved long-life testing, mean
stress effects, and size effects. Each of these three tasks address a
sepa-rate issue of concern affecting our ability to predict the fatigue life of
weidments. For example, there is concern whether linear cumulative damage
is accurate in the long-life regime. Also, mean stress effects are not
generally dealt with, and there is concern that neglecting mean stress
introduces a considerable inaccuracy in the fatigue life prediction methods. Lastly, one generally ignores the influence of the absolute size of
weld-ments, and there is increasing evidence that there is an effect of size on
the fatigue life of weldinents. These phenomena were studied experimentally,
The third category was the application of the I-P Model for total
fa-tigue life prediction to the ship details considered in this program. The
I-P model was proposed as a basis for fatigue rating of ship details, but
this task (Task 5) was deleted at the outset of the program. The I-P model
in its current state of development is summarized in Appendix A. Section 4
compares the predictions made using the Munse Fatigue Design Procedure and
the I-P Model with the experimental test results (Task 6).
The fourth category (Task 7) was a program of fatigue testing of
se-lected ship details for which inadequate fatigue test data currently exists. The results of constant amplitude testing of the selected ship details is
summarized in Section 5.
1.5 The Munse Fatigue Design Procedure (MFDP)
The Munse Fatigue Design Procedure MFDP [l-41 is an effective method of design against structural fatigue and deals with the complex geometries, the
variable load histories, and the variability in these and other factors
encountered in the fatigue design of weidments.
Figure 1-2 shows the output from the University of Illinois Fatigue
Data Bank for a mild steel double-V butt weld. The Munse method fits such
data with the basic S-N relationship shown in Fig. l-3. When stress
histories other than constant-amplitude are used, different S-N diagrams
result if the test results are plotted against the maximum stress: see Fig. l-6. The Munse method accounts for this effect by introducing a term ¿
which when multiplied by the constant amplitude fatigue strength at a given
life will predict the fatigue strength for the variable load history at the
same number of cycles: see Fig. l-7.
Similarly, the natural scatter in fatigue data shown in Fig. l-8
to-gether with the uncertainties in fabrication and stress analysis are dealt
with by the MFDP through the concept of total uncertainty.
Û2 Û2 + m2 Û2
n f s c
where, Û the total uncertainty in fatigue life.
-
íCf + in which Cf is the coefficient of variation in thefatigue life data about the S-N regression lines; and is the
error in the fatigue model (the S-N equation, including such
effects as mean stress), and the imperfections in the use of
the linear damage rule (Miner) and the Weibull distribution
approximations.
û = the uncertainty in the mean intercept of the S-N regression
lines, and includes in particular the effects of workmanship
and fabrication. A model for this uncertainty is suggested in
Section 4.3.
= measure of total uncertainty in mean stress range, including
the effects of impact and error of stress analysis and stress
determination.
Of the above mentioned sources of uncertainty, those which are best
es-timated are probably the smallest (ûf). Those which are the largest are probably the least easy to estimate (û). In modern fatigue analysis, it is
commonly believed that the greatest uncertainty is an exact knowledge of the loads to which a structure or vehicle will be subjected in service. Often the service history bears little resemblance to that which the designer
con-templates. This difficulty with application of the Munse method as with all
other design methods will require extensive field observations and
measure-ments.
Having estimated the total uncertainty in fatigue life Q, the reliabil-ity factor Rf is estimated after assuming an appropriate distribution to
characterize the load history and after specifying a desired level of
reli-ability.
The MFDP estimates the maximum allowable design stress range LSD from the weldment S-N diagram by determining the average fatigue strength at the
desired design life tSN and multiplying this value by the random load
correction factor ¿ and the reliability factor (RF):
The Munse method takes all uncertainties into account and provides a
rational framework for designing structural details to a desired level of
reliability: see Fig. l-9.
1.6 References
l-l. AISC. "Specification for the Design, Fabrication and Erection of
Structural Steel for Buildings," American Institute of Steel
Construction, Nov. 1, 1978.
Radziminski, J.B., Srinivasan, R., Moore, D., Thrasher, C. and Munse, W.H. "Fatigue Data Bank and Data Analysis Investigation," Structural Research Series No. 405, Civil Engineering Studies, University of Illinois at Urbana-Champaign, June, 1973.
The Welding Institute, Proceedings of the Conference on Fatigue of Welded Structures," July 6-9, 1970, The Welding Institute, Cambridge,
England, 1971.
Munse, W.H., Wilbur, T.W., Tellalian, M.L., Nicoll, K. and Wilson, K., "Fatigue Characterization of Fabricated Ship Details for Design,"
Ship Structure Committee, SSC-318, 1983.
Ho, N.-J. and Lawrence, F.V., Jr., "The Fatigue of Weidments
Subjected to Complex Loadings," FCP Report No. 45, College of
Engineering, University of Illinois at Urbana-Champaign, Jan. 1983.
Maddox, S.J., "A Fracture Mechanics Approach to Service Load Fatigue in Welded Structures," Welding Research International, Vol. 4, No. 2,
1974.
Schilling, C.C., Klippstein, K.H., Barsom, J.M. and Blake, C.T., "Fa-tigue of Welded Steel Bridge Members under Variable Amplitude
Table 1.1
Program Summary
line fatigue resistance.
Task Description
1. Comparison of MFC Prediction Compare prediction of Munse Criterion with other predictive
methods.
2. Long Life Testing Perform long life variable load history fatigue tests on structural
details.
3. Mean Stress Effects Check the influence of average mean stress on fatigue resistance under
variable load history.
4. Size Effect Check the influence of plate
thickness and weld size on fatigue
resistance.
5. Fatigue Rating Deleted.
6. I-P Model Application Predict long life of ship structure
through I-P Model application.
7. Fatigue Testing of Ship Selected structural details will be Structural Details fatigue tested to determine base
3
5Iigory
C
5 7 Full Penetrallon Il 12 3 4 18 '9 20Fig. l-1 Structural details provided in AISC fatigue provision [l-l].
22 23 25
AW
PP r-e i. s.---''l!.
:.
St
e UI.ullI5IpiiS1ef'iC,S -u.e!& :.-leI. r
--un. e er... .
ÒDQ seO D D-Mild Steel R
O AWButt Welds, As Welded
= Plain Plate I t
tI''
n t n ntilt
Lower Tolerance Limit-99%
-Surv%val
-- 50% Confidence Level
-95% Confidence Level:
Do---.
I Iiiititl
ti
I111111!
tI
I1111111
t I I111111
Cycles To Failure, In Thousands
Fig. l-2
Stress range versus cycle to failure for mild steel butt welds subjected to zero to tension loading. The fatigue resistance of as-welded butt weld is generally less than the fatigue resistance of plain plate which is also indicated in
this figure [l-2]. 200 I00 80 60 u, t', ti, 40 L. U) 8 C 4 2 2
4 68CC
2 4681000.
2 2 46810
4 6 810,000Log S
Log ii
Fig. 1-3
Basic S-N relationship for fatigue [1- 4].
Log
= Log Cm Log S
n
Determine Load Histogram
tC)etermine Structural Resoonse
Fatigue Data Bank
LS
Cycles
Determine Allowable Stress
(RF) ()
'Determine Load Histogram
t9etermine Structural Resoonse
Determine Notch-root
Stresses and Strains
Strain
(
Calculate Initiation and
I
Proragation Lives
NI af -n
Di=l
Np-l/C
J51AK daFig. l-4 Fatigue design method. Fatigue design method based on detail
S-N diagrams (left) such as the Munse approach compute the
design stress SD based on corrections to constant amplitude fatigue resistance for the effects of variable load history
( ) and the desired reliability (RF). Fracture mechanics based
design methods (right) deal with the local strain events at
the critical locations and provide estimates of the fatigue
crack initiation life (N1), fatigue crack propagation life (Np) or the total fatigue life (N1 + Nr).
due to detail geometry 50% N - 10
N-106
J log cycles (N)s-
50%s-
50%s-
50%Sensitivity of S QIO level of reliability (Rel)
50% reliability - 1.00 550% 90% reliability - 0.70 95% relIability - 0.60
°'
99% reliabilIty - 0.45 S
90%
SOt
95% 99%
Sensitivity of S 0to variable load history factor:
(VLH) Welbull (k-1) S - 9.105 SO' AS R--1 Beta (q-7, p-3) S - 1.385 D SOa S
- (1 0.25R)S
50% SOt 9.9Fig. l-5 A general indication of the sensitivity of the fatigue design stress to the design variables of
the joint geometry, the variable nature of the applied load, the desired level of reliability and the applied mean stress. Beta (q7. p7) S - 1.895 AS D SO% Variation in S
due to mean stress
50%
I
69.4 ksl. 31.1 ksl. (100%) (100%) 20.3 ksl. 5.5 kaI. (29%) (2 1%) 21.5 kaI. 5.8 ksl. (31%) (19%) o s min. constant amplitude S - 1.005 D 50% AS Beta (q-3, p-7) S - 2.835 AS D 50%t.
E E D' oij4c
laco
i':
\ I\
I\
I/ 'I
's 's s' s-Log S 's 's's
''s
%.s_'
5*5n
PI
¡P
Welded Soecimen St37/St52 I I I IOio
lO No. of CyclezFig. l-6 Fatigue resistance of a weldment subjected to variable
loadings [l-2].
n
Log n
Fig. l-7 Relationship between maximum stress range of variable (random) loading and equivalent constant-cycle stress range [l-4J. Shape of the Amplitud e
Distribution
25 20 5 I h Li. ll 108io
io
Log SR
s
Constant Cycle Fatigue
- C = stm Constant Cycle. E
TL
S0 S x RF S x(I)rn
nUseful Mean Life
Life At For Design
L (n)
StressLog n
Fig. l-8 Distribution of fatigue life at a given stress level {l-4].
N
Fig. l-9 Application of reliability factor to mean fatigue resistance
2. COMPARISON OF THE AVAILABLE FATIGUE LIFE PREDICTION METHODS (TASK 1)
The effect of variable loadings on the fatigue performance of welds is
generally accounted for by using cumulative damage rules. These rules
at-tempt to relate fatigue behavior under a variable loading history to the
behavior under constant amplitude loading. The Palmgren-Miner linear
cumulative damage rule (or commonly, "Miner's rule") is widely used in many
current standards and design codes. Several models for predicting weldxnent
fatigue life have been proposed based on the S-N curve for weld details and
Miner's rule.
There are essentially two types of prediction models reported, and these are summarized in Table 2.1. The first type is based on the S-N
diagrams for the actual weld details, and the Munse Fatigue Design Procedure
is in this category. The second type is based on the fracture mechanics and
the fatigue properties of laboratory specimens, and the I-P model is in this
category
-2.1 Models Based on S-N Diagrams
The S-N diagram approach is conventionally used in current practice.
Miner's rule is used for the cumulative damage calculations:
=1
(2-l)where ni is the number of cycles applied at stress range ASj in the variable loading history and N is the constant amplitude fatigue life corresponding
to AS1. While Miner's rule usually gives slightly conservative life
predic-tions, it has been found to give unconservative life predictions for certain
types of variable loading history [2-l]. Two better methods of damage
accumulation have been proposed to predict the fatigue strength of weldments.
The first method uses the Miner's rule but modifies the fatigue limit
of the constant amplitude S-N curve for the welded detail. Figure 2-1 shows two typical ways of modifying the S-N curve. One way is to extend the
example, Schilling and Klippstein [2-2] have employed an equivalent stress range of constant amplitude that produces the same fatigue damage at the
variable amplitude stress range history it replaces. As the negative
reci-procal slope of S-N curve is about three for structural steel and structural details, Schilling et al. suggested the use of the "root-mean-cube (RMC)
stress range" for welded bridge details subjected to variable amplitude
loading history.
The other way suggested in BS 5400 [2-3] is changing the S-N curve from
a slope of -1/rn to -l/(m-i-2) at l0 cycles.
The second method for improving damage accumulation is to introduce a
nonlinear damage rule. In the Joehnk and Zwerneman's nonlinear damage model
[2-41, the ratio of damage to stress range increases nonlinearly as the
stress range decreases. Effective stress ranges were defined for subcycles
first, then Miner's rule was employed to calculate the damage of subcycles. Two fatigue prediction models have been proposed to predict the fatigue
resistance of welds subjected to variable loading history using constant
amplitude S-N diagram and will be discussed below: one uses Miner's rule and
an extended S-N curve, the Munse Fatigue Design Procedure, and the other uses and empirical relationship based on test results, Gurney's model.
Munse's Fatigue Design Procedure
The Munse Fatigue Design Procedure was reviewed in Section 1.5 and can
be used as a prediction method if one considers the variation in the random
variables to approach zero. Three factors are considered in Munse Fatigue Design Procedure l-4]: (a) the mean fatigue resistance of the weld
details, (b) a "random load factor" () that is a function of variable
amplitude loading history and slope of the mean S-N curve, and (e) a
"reliability factor" (RF) (roughly the inverse of the safety factor) that is
a function of the slope of the mean S-N curve, level of reliability, and a
coefficient of variation here taken to be 1.
The maximum allowable fatigue stress range SD for welds subjected to variable loading history is obtained from the following equation:
where ESN is the constant amplitude stress range at fatigue life of N
cycles. For welds subjected to a constant amplitude stress-range (SN), the
mean fatigue life N is given by the relationship:
N
(ASN)m
(2-2)
where C and m are empirical constants obtained from a least-squares analysis
of S-N diagram data. Munse's procedure uses the extended straight S-N line
at the stress ratio RO as its basis (see Fig. 2-l) and neglects the effects
of mean stress, material properties, and residual stress.
After cycle counting, the variable load history is plotted in a stress range histogram. Mean stress level and sequence effects are regarded as
secondary effects. Since random loadings for weld details usually cannot be
determined exactly, Munse's procedure uses probability distribution
func-tions to represent the weld fatigue loading. Six probability distribution
functions are employed to represent different common variable loading
histories: beta, lognormal, Weibull, exponential, Rayleigh and a shifted
exponential distribution function. It is necessary to determine which
distribution or distributions provides the best fit to a given loading
history. The random load factor in Munse's procedure are for a desired life
and are tabulated in [l-4]. Table 7.5 in [l-4] gives coefficients to adjust
values of ¿ to other design lines. In this study, the values of random load factor have been derived for any arbitrary fatigue life and are shown in
Table 2-2.
The reliability factor is given by:
[PF(N)]û°8
)l/m Rf-1.08 (2-3)
where PF(N) is the probability of failure, ÛN is the total uncertainty for fatigue life of N cycles and r is the gamma function.
In Ref. l-3 it is suggested that this relationship can be represented
nN.
li
N_N([ll
NI
b C 2 i (2-4)where Nb = the fatigue life in blocks
= the fatigue life in cycles at maximum stress range in the block
his tory
Ni = number of cycles per block equal or exceeding Pj times the maxi-mum stress range in the block history
n = total number of cycles in a block
The parameter contained within the braces is the random load factor.
2.2 Methods Based upon Fracture Mechanics
Methods based upon fracture mechanics ignore the fatigue crack
initia-tion phase and calculate the fatigue crack propagainitia-tion life only. Maddox
[l-61
used linear fracture mechanics and Miner's rule to predict the fatigue life of welds subjected to variable loading history. Miner's rule was found to be accurate for welds under loading histories without stress interaction.Barsom
[l-61
used a single stress intensity factor parameter,root-mean-square stress intensity factor, to define the crack growth rate under
both constant and variable amplitude loadings. The root-mean-square stress
intensity factor AK , is characteristic of the load distribution and is rms
independent of the order of the cyclic load fluctuations. Hudson [2-6]
applied the root-mean-square (RMS) method for random loading history with
variable minimum load. This simple RMS approach has been shown applicable
50% Reliability RF l_00
90% Reliability RF - 0.70
95% Reliability RF 0.60
99% Reliability RF = 0.45
Gurney's Model
Gurney [2-5 J performed fatigue tests on fillet welded joints using
simple variable loading history. It was found that the logarithm of number of blocks to faílure varied linearly with the ratio of the subcycle's stress
for loading history with random sequences. The root-mean-square stresses
are defined as:
and rrnS = (S )211/2 max LN max J n=l N = (Sa. )2]l/2 min N min n=l (2-5) (2-6)
where S and S . are the maximum and minimum stress for each cycle
max min
respectively, and N is the total number of cycles for the random loading
history.
The root-mean-square stress intensity factor range is calculated from
AK = KrmS
-rms max min (2-7)
Calculation of fatigue crack propagation life is through the substitution of Eq. 2-7 into the fatigue crack propagation model, Eq. A-18.
A deterministic model for estimating the total fatigue life of welds
has been developed by the authors and is presented in Appendix A. This
model is termed the initiation-propagation (I-P) or total life model and
assumes that the total fatigue life of a weld (NT) is composed of a fatigue
crack initiation (N1) and a fatigue crack propagation period (Np) such that:
NT N1 + Np (2-8)
The initiation portion of life may be estimated using the fatigue data from strain-controlled fatigue tests on smooth specimens. The initiation
life so estimated includes a portion of life which is devoted to the
development and growth of very small cracks. The fatigue crack propagation
portion of life may be estimated using fatigue crack propagation data and an
arbitrarily assumed initiated crack length (ai) of O.01-in. in the instances in which the initial crack length is not obvious. A second alternative is
to assume that a is equal to ah the threshold crack length. In most
within a factor of 2 [1-5]. Naturally, for welds containing crack-life
defects, N1 may be very short. However, for other internal defects having
low values of Kt such as slag or porosity, N1 may be appreciable; and
neglecting N1 may be overly conservative. This is particularly the case for welds containing no discontinuities other than the weld toe. In this case
and particularly for the long life region, it is believed that the fatigue
crack initiation portion life (as defined) is very important. A detailed
discussion of the I-P model is given in Appendix A.
2.3 Comparisons of Predictions with Test Results
Table 2.1 sununarizes the prediction models discussed above. Several of
these models were used to predict the "mean fatigue lives" of welds tested
in this and other studies [2-7]. Figures 2-2 to 2-10 compare the predic-tions made by the Munse Fatigue Design Procedure, Miner's rule, Gurney's model, the RMS method, and the I-P model with actual test data for several
histories. The Munse Fatigue Design Procedure (MFDP) and the Miner's Rule
predictions in these figures differ only in that the MFDP uses a continuous
probability distribution function to model the load history while the
Miner's rule sums the actual history. The "Rainflow" counting method was
used in these comparisons. In these comparisons, the maximum stress in the
load history (5Arn or S ) is plotted against the predicted life. The
min max
effects of bending stresses were taken into account.
The Munse Fatigue Design Procedure (MFDP) provided good mean fatigue
life predictions for welds subjected to the SAE bracket history (See
Appendix C) as shown in Figs. 2-2, 2-3, and 2-5. For welds tested under the SAE transmission history (See Appendix C), unconservative predictions were
made by the MFDP (Fig. 2-4). This discrepancy might be due to means stress effects because the transmission history has a tensile mean stress while the
bracket history has only a small average mean stress. The root-mean-square method (fatigue crack propagation life only) gave conservative predictions
for all cases. It is interesting to note that the predictions made based on
S-N curves without cutoff and Miner's rule are similar to the predictions of
the MFDP. Predictions resulting from the Total Fatigue Life (I-P) model seem to agree well with the test results. Table 2-3 is a statistical
sum-mary of the departures of predicted lives from the test data as in Fig. 2-6
to Fig. 2-10.
While the agreement between the prediction methods discussed above and the two variable load histories employed in the comparison are quite good, there are histories for which all predictions methods based on linear cumu-lative damage fall short even when the very conservative assumption of an
extended S-N diagram is used [2-8]. These histories are typically very long histories in which most of the damaging cycles are near the constant
amplitude S-N diagram endurance limit. Neither the SAE bracket or
transmis-sion histories nor the edited history discussed in the next section fall into this category; consequently, this serious problem in fatigue life
prediction is not addressed by the comparison of this section nor the
experimental study of the next section.
2.4 References
2-l. Fash, J.W., "Fatigue Life Prediction for Long Load Histories," Digital
Techniques in Fatigue, S.E.E. mt. Conf., City University of London,
England, March 28-30, 1983, pp. 243-255.
2-2. Schilling, C.C. and K.H. Klippstein, "Fatigue of Steel Beams by
Simu-lated Bridge Traffic," Journal of Structural Division, Proceedings of
ASCE, Vol. 103, No. ST8, August, 1977.
2-3. BS5400: Part lO: 1982, "Steel Concrete and Composite Bridges, Code of Practice of Fatigue."
2-4. Zwerneman, F.J., "Influence of the Stress Level of Minor Cycles on
Fatigue Life of Steel Weldments," Dept. of Civil Engineering, The
University of Texas at Austin, Master Thesis, May 1983.
2-5. Gurney, T.R., "Some Fatigue Tests on Fillet Welded Joints under Simple
Variable Amplitude Loading," The Welding Institute, May 1981.
2-6. Hudson, C.M., "A Root-Mean-Square Approach for Predicting Fatigue
Crack Growth under Random Loading," ASTM STP 748, 1981, pp. 41-52.
2-7. Yung, J. -Y. and Lawrence, "A Comparison of Methods for Predicting
Weldment Fatigue Life under Variable Load Histories," FCP Report No. 117, University of Illinois at Urbana-Champaign, Feb., 1975,
2-8. Gurney, T.R., "Fatigue Test on Fillet Welded Joints to Assess the Validity of Miner's Cumulative Damage Rule," Proc. Roy. Soc., A386,
2-9. Miner, M.A., "Cuxnalative Damage in Fatigue," Journal of Applied
Basis Proposed by
S-N curve Miner [2-9]
Summary of Fatigue Life Prediction Models For Weidments Subjected to Variable Loadings
Zwerneman [2-4] Joehnk Gurney [2-5] Munse [l-4] fracture Barsom [l-7] mechanics Lawrence [l-5] Ho Table 2.1 Model (ni/Ni) = 1
ni : no. of cycles applied at
no. of cycles to failure at linear damage accumulation
ASeff ASi(ASmax/ASi)a
ASeff : effective stress range at ASmax
AS1 : stress range of subcyles
ASmax : maximum stress range
a : varies with loading history
nonlinear cumulative damage
n
p-N N [ fi (N . /N .)
'J
b c
2 ei-i ei
Nb : no. of blocks to failure
N : no. of cycles to failure at ASmax
Nei: no. of cycles per block equal to or exceeding Pi times the maximum stress in one block
SD - SN * * RF
SD : allowable maximum stress range
SN : maximum stress range in life N
probabilistic random load factor reliability factor
rms = [( A}()2/n]1/2
rms : root mean square stress intensity
factor range
fatigue crack propagation life only
NT N1 + N
NT : total Eatigue life
N1 : fatigue crack initiation life
Distribution Function Random Load Factor, ¿ beta ([r(q)r(m+qr)]/[1'(m+q)l'(q+r)]) Weibull (inN)hh'k[r(l+m/k)] -1/rn exponential (inN) [r(l+m)]*1/rn Rayleigh
(jp)l/2[r(l+m/2)]/m
lognormal shifted exponential Û : Coefficient of Variation of F Fp P Table 2.2Random Load Factors for Distribution Functions [1-4]
i m [ m!/(mn)!(2nNYn(l
-a)a
nm-n -1/rn n=0 a = a/[a-4-z(inNb)) Table 2.3Statistical Summary of the Departures of Predicted Lives from Fatigue Test Data
1/rn No. of Cases 29 29 29 13 29 1.061 1.015 0.894 0.906 1.016 cl 0.124 0.093 0.081 0.052 0.067 Fp log10 (N di ti Mean value of F; F
log10
unity nf F
valueTest'
represents the perfect agreement between the prediction and fatigue
data.
Munse' s Miner' s Curney' s RMS Method I-P Model
S-N Curve (In Air)
m
f
With Fatigue Limit
rBS 5400
J
m+2
Extended Line
-Log Nf, cycles
Fig. 2-1 Modification of S-N diagram.Io'
I
Il
GM MS4361 Steel
Fillet Welds
SAE Bracket History
IJP Failure
-: Test Results
:1-P Model
-- :
Muns&s Beta (S-N)
--- :
Miner's Rule (S-N)
RMS Method
I I i i
till
'a'
IO'
102NT,
blocks
Fig. 2-2Fatigue test results and predictions for GM MS4361 Fillet welded cruciform joints subjected to SAE Bracket history.
S1
is the minimum stress in the load history.
The Incomplete Joint penetration (IJP) is indicated by 2C.
I I I I I
titi
I I I Ititi
I I t I IIIi
102
lo'
CT 1E650-B Steel
Fillet Welds
-SAE Bracket History
.Ijp Failure
A
: Test Results
I-P Model
-- :
Munse's Beta (S-N)
--.- : Miner s Rule (S-N)
RMS Method
i I I I iIii
I I ¡ I IIIIj
-A
I I I I IlIti
4
IO'
IO'
102
IO IO4NT,
blocks
Fig. 2-3Fatigue test results and predictions for CT 1E650-B
fillet welded cruciform joints
subjected to SAE Bracket history.
Smtn is the minimum stress in the load history.
The Incomplete Joint penetration
ASTM A36 Steel
Butt Welds
SAE Transmission History
Toe Failures
A
:
Test Results
I-P Model
--: Munse's Rayleigh (S-N)
Miner's Rule (S-N)
--: RMS Method
__P
I I I111111
I I IIIIlIJ
I 1111111
D
IO'
X D<E
(J)lo'
102 IO3NT
blocks
Fig. 2-4Fatigue test results and predictions for ASTM
A 36
butt welds subjected to SAE
transmission
history
SA
is the maximum stress in the
load
history.
min
102
ASTM A588-A Steel
Butt Welds
SAE Bracket History
Toe Failure
: Test Results
I-P Model
--: Munse's Beta (S-N)
-.-: Miner's Rule (S-N)
N1, blocks
Fig. 2-5
Fatigue test results and predictions for ASTM A588-A butt welds subjected to SAE Bracket history.
S
is the minimum stress in the load history.
20
10û
lO'
I-C
<E
w
I&
102l0
102
102
I I IIIIIt
i J1111111
I II'IIII
:
Cruciform Joints
/
:GMMS436I
/
/1
-
A:CTIE65O-B
/
/
-Butt Welds
/
/
o: ASTM A588-A
/
/
: ASTM A36
/
AAA/
/
/
/
/
/.
A/
/
/
0/
/
/
/
s
/
Munses Fatigue Criterion
i
iii iii
i i iii iii
i i i it tu
/
Actual Life, blocks
Fig. 2-6 Comparison of actual and predicted fatigue life using Munse's
Fatigue Criterion (Extended S-N Curve). The dashed lines
u,
O ir4
o
-Q
/
/
Cruciform Joints
e: GM MS4361
ACT lE650-B
Butt Welds
o: ASTM A588-A
: ASTM A36
/
/
/
/
/
/
/
/
/
=.,
/
0/
A/
Miner1s Rule
-,/'
(Based on Straight
2'
S-N Curves)
ic
ij titil
i i iill iii
I I III
11102
IOIO4
IO5Actual Life, blocks
Fig. 2-7 Comparison of actual and predicted fatigue life using Miner's
Rule and the Extended S-N Curve. The dashed línes represented
102
-
I I I111111
:
Cruciform Joints
/
- : GM MS4361
=
:CTlE65O-B
/
/
Butt Welds
0:
ASTM A588-A
A:
ASTM A36
/
/
-
/
/
I
/
/
-
/
/
/
/
-
/
/
-
/
/0
/
A/A
°
Gurneys Model
/
¿
I Ii III
I I I Ili iii
i i t Ililt
Actual Life, blocks
I I I i
Ili5
¡ iIII
Fig. 2-8 Comparison of actual and predicted fatigue life using Gurney's
Model. The dashed lines represented factors of two departures
a)
-J
-o
a).4-o
-o
a) û-I I II II
I I I I I I I j I I ¡ ICruciform Joints
/
s: GM MS4361
//
/
£:CTIE65O-B
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/
/.
io3/
/
/AA
A//
-
/
A//
RMS Method
//
/
/s
/
102 ijiltilil
iIIIHII
102
io3Actual Life, blocks
Fig. 2-9 Comparison of actual and predicted fatigue life using the
RMS Method. The dashed lines represented factors of two
102
102
ìctual Life, blocks
Fig. 2-10 Comparison of actual and predicted fatigue life using the
t-P model. The dashed lines represented factors of two
3. FATIGUE TESTING OF SELECTED SHIP STRUCTURAL DETAIL
UNDER A VARIABLE SHIP BLOCK LOAD HISTORY (TASKS 2-4)
The experimental portions of this study can be divided into two parts:
The first and major effort of this study was to test a selected structural
detail under a variable load history which simulated a ship history. This
part encompassed Long Life Variable Load Testing (Task 2), Mean Stress
Ef-fects (Task 3), and Thickness Effects (Task 4). The results of the first part of the experimental program are discussed in this section. The second
part of the experimental program (Task 7) was the collection of baseline
constant amplitude fatigue data for selected ship structural details. The
results of this latter study are summarized in Section 5.
A major technical difficulty at the outset of this part of the
experi-mental program was obtaining a variable load history which simulated the
typical service history experienced by ships. Since no standard ship
his-tory was available, the first major task for the experiments described in
this section was to develop a reasonable variable load history block which
simulated the load history of a ship. This task was further complicated by the fact that typical ship histories have occasional large overloads which
cannot be contained in every block or repetition of a short history and by
the large number of small cycles which contribute little to the accumulation of fatigue damage but which enormously increase the time required for
test-ing and consequently determine whether or not the laboratory testtest-ing can be
completed in a reasonable time.
3.1 Determination of The Variable Block Load History.
The Weibull distribution was demonstrated to be an appropriate
proba-bility distribution for long-term histories through comparison with actual
data such as the SL-7 container ship history [3-l]: see Fig. 3-l.
Munse [l-4] used the 36,011 scratch SL-7 gauge measurements or records taken over four-hour periods shown in Fig. 3-1. Each measurement or short
history contained 1,920 cycles. The biggest "grand cycle" of each history
was termed an occurrence, and the 36,011 occurrences were assembled into the
histogram shown in Fig. 3-2 and fitted with a Weibull distribution (k = 1.2,
histogram for the entire history composed of 52,000 short histories containing 1,920 cycles each, or io8 cycles. Using the fitted Weibull
distribution, Munse estimated the maximum stress range expected during the
ship life of 108 cycles (S10-8) by assuming that the probability associated
with this stress range would be 1/108, that is, equal to that for the
largest occurrence. From this argument and the fitted Weibull distribution,
a maximum stress range of 235 MPa (34.11 ksi) was calculated as the maximum
stress in the 20 year ship life history for the location at which the stress
history was recorded.
The SL-7 history and Munse's Weibull distribution representation of it
was adopted for use in this study. The next problem was to create a typical
history, that is a sequence of stress ranges which represented the typical
ship experience (period of normal sea state interdispersed with storm
epi-sodes), which conformed to the overall Weibull distribution. Furthermore,
to permit long-life fatigue testing, the history had to be edited to remove
cycles which caused little fatigue damage but needlessly extended the
required testing time. It was decided to edit the history so that one
"block" would contain only 5,047 cycles and yet contain the most damaging
events in a typical one month (345,600 cycle) ship history (see Fig. 3-6). The first step was to decide which of the events in the SL-7 history
were the most damaging and which were the least damaging and could therefore
be omitted. The damage calculated for a given interval of stress range of
the SL-7 history depends upon three things: the method of summing damage (linear accumulative damage or Miner's rule was used); the assessment of damage caused by a given stress range (we used the I-P model [l-51 rather
than the extended S-N approach of Munse and this makes a big difference in
what can be omitted); and the degree of stress concentration by the weld defects to be studied (we assumed a maximum fatigue notch factor (K )
fmax
typical for Detail No. 20 as K = 4.9). fmax
The justification for adhering to the predictions of our I-P model is
that it has given reasonable estimates of weldment fatigue life under
varia-ble load histories in laboratory air [2-6]: see Fig. 2-10. A major
differ-ence between the I-P model and Munse's approach is the anticipated behavior
of the weidments in the long-life region. Fig. 3-3 shows the extended line
exaggerates the importance of the smaller stress ranges and leads to the
conclusion that they can not be deleted. The I-P model predicts that the S-N curve has a slope of about 1:10 in the long life region and consequently,
predicts a lesser importance for the smaller stress ranges: see Fig. 3-4.
We used the following strategy for editing the SL-7 history. If each cycle had an average period of 7.5 seconds as reported, a one month ship
history would consist of 345,600 cycles [3-fl. Keeping only those stress
ranges which contributed 92.8% of the total damage (estimated using the I-P model and Miner's rule, see Figs. 3-4 and 3-5) leads to the elimination of
stress range less than 68.9 MPa (10 ksi) and greater than 152 MPa (22 ksi). This decision would permit a reduction in length of the one month history
from 345,600 (total) cycles to 5,047 cycles. Fig. 3-6 shows the developed
"one-month history" which starts with a period of low stress range, 75.8 MPa (11 ksi), and gradually increases to a maximum of 145 MPa (21 ksi) during the central storm period after which the amplitude decreased to the original
11 ksi. At a testing frequency of 5 Hz, a block required about 17 minutes.
Since a 5,047 cycle block represents 345,600 cycles in service, 290 blocks
or 3.5 days of testing at 5 Hz or 10 to 15 days at lower testing frequencies are equivalent to a 108 cycle service history or 24 years of service.
The ship block load history shown in Fig. 3-6 was read into the memory of a function generator which controlled a 100 kip MTS fatigue testing
machine.
3.2 Development of a "Random" Ship Load History
At the suggestion of the advisory committee, an alternative "random"
time history was generated using a method employed by Wirsching [3-2]. To simulate a stress history from a given spectral density function, the
spec-tral density must be discretized. This operation was accomplished by
defin-ing n random frequency intervals, in the region of definition of f. The value of must be random to insure that the simulated process is a nonperiodic function, f is the midpoint of iïf1. The simulation is
con-structed by adding the n harmonic components:
n
y(t) cos (2irf t (3-1)